Introduction

The fourth industrial revolution is heralding a new era of digital transformation and associated technological advances and integrations across the globe (Anshari and Almunawar, 2022). The application of smart technologies is becoming increasingly pervasive across all industry sectors, and the implications of these technologies for HRM strategy and practice in organizations are beginning to be felt more acutely. The imperatives of digital transformations coupled with the effects of the recent global pandemic are driving sonic-speed changes within business processes and systems. These macro trends are in turn impacting HRM strategy and practices as we witness an ever-increasing use of algorithms in HRM, heralding the so-called ‘Era of Algorithmic HRM’ (AHRM) (Kim et al. 2022, p. 1). This era of AHRM will have significant implications and impacts for organizations and the efficient management of their human resources. These digital and algorithmic applications can provide efficiencies and promise increased productivity by releasing employees from more routine tasks. These applications will also require changes in organizational management systems and operations which in turn will necessitate upskilling and change management strategies to ensure the effective implementation of these applications.

To address these trends and advances we conduct a multidisciplinary synthesis of the concepts related to algorithms which results in a General Framework for Algorithmic Decision-Making. This framework then assists us in focusing on algorithmic HRM applications and mapping the evolution of AHRM and allows us to present a Framework for Algorithmic AHRM Tools. This provides conceptual clarity and distinguishes between automated and augmented HR decision-making and its implications for strategic HRM.

The link between digital transformation and strategy and SHRM is undeniable where both human capital and technologies “are perceived as assets and competencies owned or controlled by the firm that generate value for the business in achieving competitive advantage’ (Fenech et al. 2019, p. 166). Fenech et al. (2019) explore the digital transformation of HRM through the lens of the resource-based view (RBV) and propose that organizations represent “a collection of unique resources and capabilities that provides the basis for its (corporate and HRM) strategies and is the primary source of its (financial) returns” (Hanson et al. 2008, p. 23). Their particular focus is on the functions of HR planning, performance and reward management, training and development, employee relations, and health and safety. The study emphasizes the impact that this is having on HR professionals and practitioners. Boon et al. (2018), however, suggest that the majority of SHRM researchers focus on more macro-perspectives such as ‘the investment in human capital to increase firm performance, by using systems and practices aimed at developing and managing an organization’s human capital’ (p. 36), which require more sophisticated use of AHRM technologies.

From a slightly different perspective, Einola and Khoreva (2022), refer to “low-status” automation, and “high-status” augmentation when investigating the concept of digitalization in the workplace which represents the adoption and application of emergent and sophisticated technologies. Their focus is on the relationship between and co-existence of humans and artificial intelligence (AI) in the workplace. Similarly, Malik et al. (2023) explored the configuration of a digitalized HR ecosystem of AI-assisted HRM applications and HR platforms by gaining insights from a multinational company, with particular emphasis on the employee experience and employee engagement. They claim that “there is a limited theoretical basis for understanding how AI-assisted HRM infrastructure fits into an organization’s broader ecosystem and how firms design and implement an HR ecosystem and a configuration of digitalized AI applications to cater to the firm’s digital, human and physical aspects of the work environment” (Malik et al. 2023a, p.1).

Digital HRM or eHRM are terms that have been commonly used to group electronic and digital applications in HRM which has led to what Meijerink et al. (2021) argue to be a lack of construct clarity. They argue that several key terms, concepts, and topics are loosely aggregated under this umbrella term of ‘digital HRM’. These include HRM analytics, HRM algorithms, online platforms, artificial intelligence, and big data used in HRM decision-making and employee recruitment, tracking, surveillance, and workforce planning. The almost ‘splatter gun’ effect of these topics and thin associations has led to this lack of construct clarity and unifying research around these topics (Meijerink et al. 2021). As a result, these authors adopt the term “Algorithmic Human Resource Management”, hereon abbreviated as AHRM, and define the term as, “the use of software algorithms that operate on the basis of digital data to augment HR-related decisions and/or to automate HRM activities” (Meijerink et al. 2021, p. 2547). The term has three main features: the generation and use of big data; the processing of digital data using software algorithms; and partial or full automation of decision-making in HR.

This study has dual objectives. The first is to advance conceptual clarity about the term AHRM. Addressing this objective then allows us to address the second objective which is to map the evolution of AHRM across 23 years (2000 to 2022).

This paper will now provide an overview of previous reviews of various algorithmic applications in HRM with ten of these reviews and their respective findings presented and summarized in Table 1. This is followed by a brief historical perspective on algorithms more generally before offering a theoretical framework for AHRM, which synthesizes Meijerink et al.’s (2021) definition of “algorithmic HRM” with the latest available taxonomy of AI (Herrmann, 2022). This resulting framework provides an opportunity to explore the breadth and scope of technologies being applied in organizations, and how they are impacting the management of human resources through HRM strategies and functions. We then describe our science mapping methodology, which is followed by a presentation of the results and a discussion of the findings. We conclude the paper with the presentation of a General Framework for Algorithmic HRM Tools which advances conceptual clarity for this exciting and emergent field. This is followed by addressing the limitations of the study and noting areas for future research.

Table 1 Summary of previous related reviews (algorithms in HRM).
Full size table

Previous reviews of AI applications in Hrm

In this section, we present a series of reviews that analyze the use of algorithms in HRM as published in scholarly journals and proceedings. The first study we explore is a retrospective review of research on technology and HRM over a sixty-year period (1961 to 2019), based on technology-related articles published in the Human Resource Management journal (Kim et al. 2021). The study noted that while the frequency of published papers fluctuated over the sixty-year period, there were distinct peaks in publications of technology-related articles. With respect to this, they identified two landmark points in time that aligned with the advent of personal computers (1977) and the emergence of consumer internet services (1997). The study encompasses three time periods (1961–1976; 1977–1996; 1997–2019) and provides a retrospective map of technological developments in the field of HRM. They identified three main thematic streams across all three time periods: the impact of technology on jobs and organizations; the use of technology to support HR practices and HR decision-making; and the management of technology workers.

The study not only provides a useful historical preamble to the summary of the identified reviews of AI in HRM summarized in Table 1 but also provides historical continuity with this research which covers the period between 2000 to 2022. The data sources, search terms, and thematic analysis are very different for each of the studies, and we do not claim a level of similarity or comparability in this respect. Nonetheless, together they demonstrate the emerging interest in algorithmic applications in the HRM field.

We now provide an overview of the relevant reviews published between 2019 to 2023 which have examined AI applications in HRM and are listed in Table 1. Most of these studies refer specifically to AI whilst others look at machine learning, digital HRM, blockchain, and big data. All have been published relatively recently with the majority in 2021. Each of the reviews has a unique focus and uses different methodologies, search terms, databases, and data sources, resulting in varying findings across different time ranges (six to twenty-year time ranges). Nonetheless, the summary of these reviews provides a useful collection and foundation for further studies, as well as providing an evidence base for the emerging body of multidisciplinary literature on this topic. The studies are briefly summarized in chronological order of publication in Table 1 below.

Four of these articles explore the use and applications of AI technologies in HRM in general (Jatoba et al. 2019; Palos-Sanchez et al., 2022; Priya and Sinha, 2021; Qamar et al. 2021). These cover a variety of HR functionalities and applications related to the employment cycle and increasing efficiencies in HR professional practices. Two articles paid particular attention to the application of blockchain technology to HR functions. Saif and Islam (2022) focused on blockchain in HRM more generally, finding two key theoretical themes that emerged in relation to blockchain: employee-systems interactions and a blockchain framework for HR, which if implemented may impact daily HR workflows and recruitment, (talent identification, authentications of qualifications and employment histories). In contrast, Mishra and Venkatesan (2021) took a more focused exploration of the use of blockchain for detecting candidate and resume fraud in the recruitment process. The remaining five articles focused on AI applications in particular HR functionalities. For example, Votto et al. (2021) emphasized the application of AI in human resource information systems with reference to the following HR functions: recruitment, performance evaluation, remuneration, and training and development. Garg et al. (2022) focused on six machine learning objectives and the applications of these to HRM. The remaining three were very specialized, with Verma et al. (2021) exploring the use of big data-driven HR practices; and Vrontis et al. (2021) investigating smart technologies, including robotics and associated automation and HRM. Di Vaio et al. (2020) focused on AI and sustainable development business models, which included HR-related models and approaches. Together the articles evidence the increasing interest by researchers in AI technologies and their applications to HRM systems, functions, and strategies.

Given the literature reviewed and the stated dual aims of this study, the following research question has been posited to assist in addressing the second aim:

RQ1: What are the dominant, emerging, and state-of-the-art themes in published research from 2000 to 2022 that examine organizational tools for algorithmic HRM (AHRM)?

Figure 1 below provides an overview of the structure of the study and aligns the overarching aims and empirical research question.

Fig. 1
figure 1

Overview of Research Aims, Questions, and Design.

Full size image

The paper now addresses the first research aim by presenting a brief historical perspective of algorithms more generally, before suggesting a general framework that synthesizes Meijerink et al.’s (2021) definition of AHRM with the latest available taxonomy of AI (Herrmann, 2022). This framework and the conceptual clarification provide the theoretical basis for addressing our second research aim. This is followed by a description of the science mapping methodology undertaken in this research that addresses our central research question. The results and findings are then discussed before acknowledging the limitations of the study, ideas for future research, and the conclusion.

Algorithmic HRM: a theoretical framework

We now discuss the origins and problematic definition of the term “algorithm”. It is a term widely used and ill-defined with overlapping and inconsistent uses across disciplines.

Origins of “Algorithm”

The definition of an “algorithm” originates from algorithm theory in mathematics, using models of well-defined problems that are solved in finite steps by abstract or real machines (Sanders, 2009). The restriction to “finite steps” is nontrivial and goes back to Alan Turing’s (1936) seminal paper before computers existed, which proved that not every problem may have an algorithm with finite steps.

With the advent of computer science, various definitions of algorithms have been provided for the solution of problems by computer software or robots (Blass and Gurevich, 2006; Knuth, 1997; Moschovakis, 2001). Many of these definitions have intertwined the generic problem-solving steps of an algorithm with its specific implementation in code on a particular machine. More recent definitions in computer science have been seeking to separate algorithmic definitions from their machine implementation (Cormen et al. 2022; Yanofsky, 2011). However, there is no unified definition of “algorithm” as a computer science term (Kroll et al. 2017).

Defining “algorithms”

This confusion is compounded by a new type of “learning algorithms” in computer science which emerged in this century and is often referred to as “algorithmic artificial intelligence” (Herrmann and Masawi, 2022). Both terms incorporate the AI technologies of machine learning, deep learning, and big data in an approach that predicts results from historical data (Mittelstadt et al. 2016). “We will soon no longer need (or wish) to provide algorithms with hard-coded hints about how to solve problems. Instead, algorithms will be provided with some basic tools …, and then left to construct for themselves tools” (Tutt, 2017, p. 100). A recent example is “generative AI” (Gozalo-Brizuela and Garrido-Merchan, 2023), which can write computer code, literature reviews, essays, music, and poems (van Dis et al. 2023).

Another recent development for algorithmic definitions is from the social sciences, whereby perceptions of algorithms are often constructed by their “public relations”, in which they have “both a public-facing identity and new promotional discourses that depict them as efficient, valuable, powerful, and objective” (Sandvig, 2015). Indeed, the rise of “social algorithms” (Lazer, 2015) is increasingly leading toward an “algorithmic culture” (Roberge and Seyfert, 2016). It is in this social and cultural context that Hill (2016, p. 36) argues that “any procedure or decision process, however ill-defined, can be called an ‘algorithm’ in the press and in public discourse”. Seaver (2017) describes this situation from an ethnographic perspective as “terminological anxiety”. Specifically, in the HRM context, the notion of an “algorithmic boss” has captured the imagination of practitioners, researchers, and the public alike (Adams-Prassl, 2019) in what is often associated in a journal’s keywords with the term “algorithmic management” (Malik et al. 2020; McDonnell et al. 2021; Meijerink and Bondarouk, 2021; Waldkirch et al. 2021).

To avoid any confusion around the term “algorithm”, we synthesized Meijerink et al.’s (2021) definition of AHRM as above with a taxonomy of AI which was developed through a systematic science mapping methodology (Herrmann, 2022). This synthesis is depicted in Fig. 2 as a General Framework for Algorithmic Decision-Making.

Fig. 2
figure 2

A General Framework for Algorithmic Decision-Making. [adapted from Herrmann (2022), the oval sizes are conceptual only].

Full size image

The framework is contextualized by algorithmic theory and within that, computer algorithms. Automated decision-making involves prescriptive analytics, which are enabled by robotics and autonomous systems (e.g., drones, robots, or self-driving cars); robotic process automation (i.e., automating repetitive tasks such as filling out forms or transferring data between systems); and metaheuristic technologies for planning and optimization, such as scheduling employees’ shifts, ensuring that all shifts are covered while considering each employee’s availability and preferences. Augmented decision-making revolves around descriptive, diagnostic, and predictive analytics with managers in the decision loop. The main technologies utilized here stem from the discipline of data science and are generally referred to as “algorithmic AI”: machine learning, deep learning, and big data. These technologies learn how to make decisions from existing data (or in the case of big data, they learn from vast amounts of data). For example, analyzing employee performance data and predicting staff turnover down to the individual level.

The terms “data science” and “analytics” are often conflated (Hamutcu and Fayyad, 2020). For our framework in Fig. 2, we define data science in terms of its interdisciplinary grounding in statistics, database management, machine learning, and information and communication technology (Van Dyk et al. 2015). Analytics is the bridge between data science and business (Nelson, 2021) by providing actionable strategic knowledge (Das, 2014).

The vertical dimension of the framework in Fig. 2 depicts a hierarchy of analytics whereby analytics of a higher order rely on analytics lower in the hierarchy as follows (Herrmann, 2022; Jyoti, 2020; Weber and Kinkela, 2020):

  1. 1.

    Descriptive analytics can be used to understand the past through statistical ratios, metrics, and reports on administrative HR data (Margherita, 2022).

  2. 2.

    Diagnostic analytics adds business intelligence for identifying potential cause-and-effect relationships in historical data through drill-down facilities, pivot-table analyses, visualizations, dashboards, and benchmarking (Zohuri and Moghaddam, 2020).

  3. 3.

    Predictive analytics is the cross-over point into algorithmic AI by providing forecasts and simulation from historical HR process data through machine learning, or its subdiscipline of deep learning (Song and Wu, 2021). Big data goes a step further beyond past process data by considering more diverse and “live”/real-time processes, such as social media (Cameron and Herrmann, 2023). The combination of big data and deep learning has most recently given rise to the field of generative AI, which produces new artifacts from a prompt (Gozalo-Brizuela and Garrido-Merchan, 2023). Such prompts are increasingly multi-modal and combine text, images, videos, and audio (Ray, 2023). Since its launch on 30 November 2022, ChatGPT has become the most prominent generative AI tool to date (McKinsey, 2023). In AHRM, it has promising applications in recruitment, employee training, and organizational communication (Rane, 2023), or more specifically in resume screening or employee onboarding (Korzynski et al. 2023; Carvalho and Ivanov, 2023).

  4. 4.

    Prescriptive analytics uses symbolic, algorithmic, or metaheuristic paradigms of AI to either recommend the best course of action, such as expert systems with HR managers in the decision loop (Bohlouli et al. 2017); or fully automate simple and repetitive HR administration tasks, such as robotic process automation (Mohamed et al. 2022). The algorithmic and metaheuristic AI paradigms both use historical data, but they are distinct. The use of the term “algorithm” in AHRM metaheuristics is an inheritance from its parent discipline of computer algorithms (Chen and Xu, 2022), whereas algorithmic AI specifically relates to machine learning (including its subdiscipline of deep learning) and big data. Symbolic approaches are knowledge-management or rules-based rather than based on historical data. However, paradigmatic combinations are emerging in a recent development that has become known as “hybrid AI” to leverage the strengths and overcome the limitations of each paradigm (Rajaee et al., 2020). For example, symbolic and algorithmic AI paradigms are increasingly combined to enhance the ethics of AHRM in terms of its ‘fairness, model explainability, and accountability’ (Barredo Arrieta et al. 2020). Algorithmic AI is also being integrated into various metaheuristic AHRM solutions to improve overall algorithmic performance (Garg et al. 2022). To cover the third permutation of AI paradigms, the synergy of symbolic and metaheuristic AI has been applied to AHRM scheduling problems (Fazel Zarandi et al. 2020).

The General Framework for Algorithmic Decision-Making informs the search strategy of the review employed in this study.

Method

The General Framework for Algorithmic Decision-Making enabled a theory-led research method by informing our search strategy on the inclusion and exclusion criteria for identifying the relevant AHRM literature. The identified literature was then analyzed through a scientometric procedure, referred to as science mapping (Santana and Cobo, 2020). The procedure uses statistical cluster analyses of bibliometric data to construct cross-sectional and longitudinal maps (Herrmann 2023a, 2023b). This methodological approach is in line with other recent AHRM studies (Mohammad Saif and Islam, 2022; Palos-Sánchez et al. 2022). However, a novel contribution of our research is that we develop an evolutionary map of the AHRM field across four periods in this century. None of the earlier AHRM studies mapped the literature longitudinally to provide an evolutionary thematic analysis. SciMAT Version 1.1.05 was chosen as the science mapping software for its strong pre-processing and evolutionary mapping capabilities over eight alternative mapping tools (Moral-Muñoz et al. 2020).

The selection of databases in most bibliographic research comes down to a choice between Scopus, Web of Science (WoS), or Google Scholar (Ruiz-Real et al. 2020). Their multidisciplinary coverage makes them all suitable for AHRM (Martín-Martín et al. 2018), although they show different search results (Cobo et al. 2011). However, Scopus covers a wider range of journals than WoS (Glänzel and Moed, 2013), and the majority of AI-related articles outside the WoS database can be found in Scopus (Ruiz-Real et al. 2020). Google Scholar includes much nonacademic, “grey” literature and does not provide a bulk export facility for bibliographic mapping (Silber-Varod et al. 2016), which is why Scopus was the chosen database for this study. A synthesis of Meijerink et al.’s (2021) definition of AHRM and Herrmann’s (2022) taxonomy of AI resulted in the General Framework for Algorithmic Decision Making (refer to Fig. 2), which assisted in framing the following Boolean search term (in Scopus syntax for replicability) with its components derived from Fig. 2:

((AUTHKEY (“human resource” OR “personnel”) AND TITLE-ABS-KEY-AUTH (“decision support” OR “expert system” OR “robotic process automation” OR “data science” OR “business intelligence” OR “machine learning” OR “deep learning” OR “big data” OR algorithm* OR “algorithmic manage*“ OR metaheuristic* OR autonom*))) AND (PUBYEAR > 1999) AND (LIMIT-TO (SRCTYPE, “p”) OR LIMIT-TO (SRCTYPE, “j”) OR LIMIT-TO (SRCTYPE, “k”) OR LIMIT-TO (SRCTYPE, “b”)) AND (LIMIT-TO (DOCTYPE, “cp”) OR LIMIT-TO (DOCTYPE, “ar”) OR LIMIT-TO (DOCTYPE, “ch”) OR LIMIT-TO (DOCTYPE, “bk”)) AND (EXCLUDE (PUBYEAR, 2023)) AND (LIMIT-TO (LANGUAGE, “English”))

The search was run on 17 May 2023 and restricted publications to peer-reviewed journals, books, chapters, conference papers, and proceedings to obtain high-quality documents. To reduce false positives outside the AHRM area, indexing errors by Scopus were avoided by specifically requiring the terms “human resource” or “personnel” in the author keywords. Only articles since the year 2000 were considered, as from then advances in computer hardware, new optimization algorithms, the advent of open-source libraries, and publicly available data sets for the development of models drove algorithmic AI into the mainstream (Dosilovic et al. 2018). A total of 1675 AHRM publications were identified. Most were journal articles (58%), followed by conference papers and proceedings (38%), and books and chapters (4%). Their titles, abstracts, and keywords were verified for the most cited 100 publications to reduce the risk of contamination by false positives. As Scopus is error-prone to incorrect indexing of preprints, duplicate publications, and incorrect indexing of references (Franceschini et al. 2016), a check for duplicates was conducted, which found three that had to be eliminated, resulting in 1672 final publications for science mapping.

Results

The results are now presented in various charts. Figure 3 depicts the frequency of publications across the period between 2000–2022. Publication counts were relatively low from 2000 to 2006, with less than twenty new publications generated per year. From 2006 to 2019 there was a steady increase in yearly publications. In 2020, publication growth doubled as compared to 2018/2019. This explosion of publications was driven by applications in HR scheduling, allocation, and recruitment.

Fig. 3
figure 3

AHRM publications by year (2000 – 2022).

Full size image

Figure 4 shows the spread of the 1672 publications across several Scopus subject areas. We found that AHRM is a multidisciplinary field with more contributions from the fields of computer science and engineering than from business, management, and accounting.

Fig. 4
figure 4

AHRM publications by Scopus subject area.

Full size image

Considering that the origins of algorithms are in computer science as per Fig. 2, it is not surprising that the majority of publications are published in computer science journals. Interestingly, the Saif and Islam (2021) study into blockchain and HRM found a similar breakdown with computer science (25.4%), engineering (16.2%), and business, management, and accounting (9.9%).

We used co-occurrences of the publications’ keywords for mapping the evolution of AHRM. A minimum threshold of two AHRM occurrences was applied across the title, abstract, and keywords to filter out “noise” caused by publications only peripheral to AHRM. Then, articles were clustered into themes for each period. Solid lines across periods in Fig. 5 below show linked themes sharing at least two keywords.

Fig. 5
figure 5

Evolutionary map of AHRM since the year 2000 (the sphere size is proportional to the h-index of publication clusters).

Full size image

Dashed lines in Fig. 5 denote themes sharing only one keyword. The thickness of solid or dashed lines is proportional to the strength of a relationship between themes. When solid and dotted lines are traced, the trace defines a thematic area, which depicts how keywords have developed into other themes. Thematic areas are shown in Fig. 5 through blue and amber colors. It can be seen in the blue thematic area that decision theory, learning and development, and data science had an early influence on the evolution of AHRM. Over time, their thematic linkages extended to the fields of workforce planning, allocation, and scheduling. Algorithmic AI and metaheuristics have been the major algorithmic technologies impacting this outcome. Augmented decisions prevail in the decision-making processes relating to the blue area although some are automated (Karim et al. 2022; Patalas-Maliszewska et al. 2021; Vrontis et al. 2022). Conversely, Fig. 4 shows that employee tracking in the amber thematic area has strong thematic linkages with Industry 4.0 and automation in recruitment (Chatterjee et al. 2021; Gaur et al. 2019; Langer et al. 2021; van den Broek et al. 2021; Yang et al. 2019; Yu et al. 2021).

The legend in Fig. 5 shows three categories of themes:

  • AHRM technologies: data science, algorithmic AI, metaheuristics, and tracking.

  • Specialized applications: learning and development, workforce planning, allocation and scheduling, and recruitment.

  • Multiple and integrated contexts: decision theory, augmented decisions, and industry 4.0 automation.

The first two categories are self-evident, but the third category requires an inspection of the internal structure of its associated thematic clusters. The internal cluster structures are depicted as a graph with themes as nodes in Figs. 6–8. The central theme of a cluster gives the cluster its name. The thickness of their connecting lines is proportional to the strength of an association between themes. The size of the spheres is proportional to the number of articles in the cluster.

Fig. 6
figure 6

Internal structure of the DECISION-THEORY cluster.

Full size image
Fig. 7
figure 7

Internal structure of the AUGMENTED-DECISIONS cluster.

Full size image
Fig. 8
figure 8

Internal structure of the INDUSTRY-4.0-AUTOMATION cluster.

Full size image

The DECISION-THEORY cluster in the 2000—2012 period shows its interdisciplinary nature in AHRM through a dense network of technical themes, such as ARTIFICIAL INTELLIGENCE, ROUGH SET THEORY, FUZZY LOGIC, etc. This suggests that decision-making in AHRM often requires the integration of insights from various technical fields. Two dominant AHRM applications are depicted: recruitment, and allocation and scheduling. The connecting line between these two applications signifies that the literature sometimes discusses them in integrated contexts of HRM problems (Huang et al. 2009; Malinowski et al. 2008; Shahhosseini and Sebt, 2011). Emerging themes in that period, such as ENTERPRISE RESOURCE MANAGEMENT, HIERARCHICAL SYSTEMS, and MANAGEMENT INFORMATION SYSTEMS, indicate the expanding scope of decision-theoretic applications in AHRM.

The AUGMENTED-DECISIONS cluster in the 2013—2017 period also shows a dense network of technical themes, and four dominant AHRM applications: recruitment, allocation and scheduling, risk, and performance management. As was the case for the DECISION-THEORY cluster, recruitment, allocation, and scheduling are sometimes discussed in integrated contexts in this period (Kim and Mehrotra, 2015). In this cluster, performance management (Chaturvedi and Joshi, 2017) and risk management (Mishra and Lama, 2016) join such integrated contexts. As before in the previous period, strong interdisciplinary technology influences are evident with METAHEURISTICS being the most influential. AHRM emerges in the context of HR information systems (HRIS).

The INDUSTRY–4.0-AUTOMATION cluster in the 2018–2020 period shows a network with lower density than the previous two clusters. In scientometrics terms, a lower density indicates a less developed cluster (Fosso Wamba et al. 2021). Its impact on the AHRM corpus with a h-index of 10 is substantially lower than for the DECISION-THEORY and AUGMENTED-DECISIONS clusters, which have h-indexes of 25 and 22, respectively.

Performance management is a major AHRM application in this cluster (El-Kassar and Singh, 2019; Kun-fa et al. 2019). Other HRM aspects associated with industrial automation, also feature here: PERSONNEL-SECURITY, SAFETY-ENGINEERING, EMBEDDED-SYSTEMS, SOCIO-ECONOMIC-SYSTEMS, HR-PRACTICES, and HR-MANAGEMENT-SYSTEMS.

Findings and discussion

Our extensive science mapping of the relevant literature from 2000–2022 reveals four key findings that suggest possible practical recommendations for future AHRM researchers and HR practitioners. First, we found that confusion remains amongst researchers (and presumably HRM professionals as well) about the nomenclature associated with terms such as artificial intelligence, machine learning, and big data, with HRM-specific umbrella terms such as AHRM, digital HRM, HR analytics, and e-HRM being used interchangeably (Meijerink et al. 2021). Such merging of the different but interlinked technologies and associated umbrella terms only serves to illustrate the lack of precision in both understanding and industry applications. This might be addressed more adequately through ongoing dialogues between the technical and HRM researchers and professionals. As Meijerink et al. (2021) have suggested, a broad definition of AHRM should include all of these components, some of which might be more applicable to particular HRM activities than others, and that all may be relevant in an integrated manner to ensure the effectiveness and accountability of strategic HRM (SHRM).

The second finding, as illustrated in Table 1 and Fig. 4, points to the large majority of articles being published in non-HRM journals, principally in the disciplines of computer science and engineering. Only 11% of articles were published in general business or more specific HRM journals. This suggests that there is only limited interest or recognition by global HRM researchers of the present and future importance of employing AI in HRM strategies, plans, and processes. The evolutionary map in Fig. 4 provides some confirmation of this. It also suggests that a multi-disciplinary approach to research and practice in this area might be fruitful in the future, including computer science and engineering researchers (inter alia) with more focused HRM scholars, in order to combine their expertise in both theoretical and practical domains. This novel approach would have a synergistic effect, as all disciplines could learn more about each other’s perspectives and thereby develop holistic and developmental protocols and systems. It could also help to demystify AHRM for HRM researchers and professionals, decrease their trepidation at employing the technologies on the one hand, and improve computer science and engineering experts’ knowledge and understanding of the complexities and challenges of developing HRM strategies, plans, and processes on the other hand.

The third finding of the study is that the majority of the articles analyzed focus on arguably what would be considered non-strategic HRM functions, such as employee scheduling and rostering, recruitment, selection, and even learning and development (refer to the specialized applications in Fig. 5), although it can be argued that these HRM functions contain some strategic elements. For example, rostering may represent broad employee deployment plans in concert with SHRM strategies; recruitment and selection might focus on human capital attraction and retention imperatives; and learning and development programs may result in higher skills and competency levels to meet broader organizational imperatives. Yet, the absence of a clear focus on using these technologies to capture and analyze data which assists organizations to maximize their human capital strategies and plans (Gallo and Thompson, 2000, p. 241) is at first glance concerning. Exceptions are workforce planning specifically and HRM analytics more generally, which are essential components of effective SHRM, and by inference, AHRM.

Other exceptions were the multiple and integrated contexts found in Figs. 5–8. Evidence for a more strategic integration was found to be most consistent for recruitment, allocation, and scheduling. Indeed, most research articles involve single objective HRM applications from a “tools view” perspective (Garg et al. 2021; Kim et al. 2021). Some examples are the predictability of workforce turnover (Willard-Grace et al. 2019; Chakraborty et al. 2021), employee engagement (Al Mehrzi and Singh, 2016; Pitt et al. 2018), learning and development needs (Chen et al. 2007; Wang et al. 2015;), career progression modeling (Li et al. 2017; Zhao et al. 2015), performance management (Aktepe and Ersoz, 2012; Gui et al. 2014), or team player dynamics (Masuda et al. 2017; Schönig et al. 2018). More complex multi-objective HRM goals have also been researched (Mammadova and Jabrayilova, 2015; Rahmanniyay and Yu, 2019; Schönig et al. 2018)), but they remain in the non-strategic HRM domain.

Lastly, we find the recruitment function is the most automated and hence, prescriptive AHRM “tool” application of the specialized AHRM applications in Fig. 5. The other specialized AHRM applications involve HRM personnel in the decision loop for augmented decision-making: allocation and scheduling, learning and development, and workforce planning. All of the algorithmic technologies in Fig. 2 have been deployed in every one of the specialized applications, except that metaheuristics are an emerging AHRM technology in recruitment (as well as for more complex, multi-objective optimizations in other HRM functions). Table 2 summarizes these findings for specialized AHRM applications.

Table 2 Findings for Specialized AHRM Applications from data-driven science mapping.
Full size table

The findings in Table 2 were derived from a data-driven science mapping procedure. We acknowledge that data-driven and theory-driven perspectives should be mutually reinforcing (Maass et al., 2018). Therefore, we integrated the findings from Table 2 with our theory-driven General Framework for Algorithmic Decision-Making from Fig. 2. This integration is shown in Fig. 9 as a General Framework for AHRM Tools.

Fig. 9
figure 9

General Framework for Algorithmic HRM Tools.

Full size image

The blue area in Fig. 9 encapsulates technologies and applications that relate to augmented decisions with humans involved in the decision-making loop. The orange area depicts automated decision domains with little or no human involvement. Algorithmic recruitment involves both augmented and automated modes of decision-making tools. However, it is more automated to date than other specialized AHRM applications, such as workforce planning, allocation and scheduling, and learning and development. These applications have traditionally been supported by data-scientific statistical modeling as well as business intelligence tools for visualizations in diagnostic analytics of the data. Increasingly, algorithmic AI supports these applications through predictive analytics in machine and deep learning, as well as big data tools.

Conclusion

Algorithmic HRM is an emergent field of inquiry, a field currently plagued by an array of umbrella terms and a lack of cohesion in relation to what is being researched and explored in HRM research under these conceptual umbrella terms (e.g., “Digital HRM”, “artificial intelligence applications in HR”, “algorithms”, “HR analytics”). This study makes a theoretical contribution to the field by synthesizing Meijerink et al.’s, (2021) definition of “algorithmic HRM” with the latest available taxonomy of AI (Herrmann, 2022) to develop a General Framework for Algorithmic Decision-Making that provides a signpost for exploring the breadth and scope of algorithmic technologies being applied in SHRM initiatives and HR practices.

The findings showcase the largest scale review of extant research in algorithmic HRM to date with a total of 1672 AHRM publications identified through science mapping. We expose the multidisciplinary nature of this emergent field of inquiry, with almost a quarter of the literature identified published in the field of Computer Science (24%), followed by Engineering (15%) and then Business and Management (11%). This has implications for HRM researchers in terms of a less blinkered and more multi-disciplinary perspective toward research that relates to algorithmic HRM including its use of AI technologies and big data for decision-making and automation in HRM applications.

The findings also cluster the extant published research into three main categories: AHRM technologies, specialized HRM applications, and multiple and integrated HRM contexts. The synthesis of these categories involved a deep inspection/analysis of the internal structure of HRM contexts and culminated in the General Framework for Algorithmic HRM Tools presented in Fig. 9. This provides conceptual clarity and distinguishes between automated and augmented HR decision-making. Findings point to current research which focuses on specialized applications for HR functions such as learning and development, workforce planning, allocation and scheduling, and recruitment; but lacks emphasis on more integrative strategic HRM contexts.

Many of the articles included in this paper have revealed both benefits and considerable challenges in the use of AHRM applications in SHRM. Key among the benefits is the increased capability of HRM strategists and planners to collect, collate and analyze large amounts of HR data in order to better prepare their organizations for present and likely future human capital demands, through systematic analysis (Saif and Islam, 2022) and data mining (Priya and Sinha, 2021); to enable sustainable business development and stimulate innovation (Verma et al. 2021); and to more effectively link SHRM to business planning (this article). As Priya and Sinha (2021) summarize it:

“Using digital technologies in HR will lead to improvement in organizational performance through talent-related decisions, forecast workforce requirements, optimizing talent through development and planning. It will also enable HR to help an organization to achieve corporate goals through informed decision-making. Additionally, it enables managing employees through recruitment, training, employee satisfaction, productivity, and assigning of tasks as per the qualification. It also helps in identifying the reason for attrition and identifying high-value employees for leaving” (p. 4373).

Conversely, with respect to the challenges faced, research focused on micro-HRM functions and activities such as recruitment and selection, learning and development, remuneration and performance management (for example, Mishra and Venkatesan, 2021; Qamar et al., 2021; Votto et al. 2021) fails to provide a holistic perspective on the role of AHRM in enhancing SHRM. It could encourage HR practitioners to only focus on the abrogation of routine and administrative components of their roles, at the expense of more important strategic opportunities. Two recent articles by Nankervis et al. (2019) and Nankervis and Cameron (2022) also question whether Australian HR professionals are aware of the potential strategic capability of AHRM and whether they have the necessary capabilities and competencies to take advantage of the opportunities offered by AHRM. In a different national context, Garg et al.’s (2022) review of the literature, argues that, at least with respect to machine learning (ML), “HRM has embraced ML, albeit it is at a nascent stage and is receiving attention largely from technology-oriented researchers. ML applications are strongest in the areas of recruitment and performance management and the use of decision trees and text-mining algorithms for classification dominate all functions of HRM. For complex processes, ML applications are still at an early stage; requiring HR experts and ML specialists to work together” (p.1590).

There are several implications resulting from this study. In order to fully embrace the benefits and face the challenges of AHRM, both HR researchers and HR professionals will need to adopt new mindsets and new ways of working within multidisciplinary teams. HR researchers will need to form multidisciplinary research teams to address research issues related to AHRM and its impact on HRM strategy and HRM functionalities. This will provide broader perspectives and diverse discipline-specific knowledge bases on the implementation, application, and consequences of AHRM adoption. For example, the ethical implications of AHRM tools and applications must be part of the research agenda for these multidisciplinary teams. The subsequent research can then assist in advancing SHRM theory and contribute to the evolving field and literature on AHRM. In turn, this research can inform HR practice and HR professional education.

This study also has practical implications for those stakeholders charged with educating and professionally developing HR professionals. The educational institutions that run HRM professional education courses and degrees and HR professional bodies play a major role in educating current and future HR professionals. University degrees with HR professional education content will need to encompass AHRM and provide students with the theories, knowledge, and skills to best inform their professional practice so that they are better equipped to operate in an AHRM era. Professional bodies need to serve their members and provide continuing professional development opportunities for learning new skills and competencies which will better equip them to grasp these new opportunities for significantly enhancing their SHRM capabilities without simultaneously diluting their human relations skills.

Limitations and future research

The paper provides some valuable insights for both organizational and HR researchers and HR practitioners alike. However, the study has limitations that need to be acknowledged. Firstly, we acknowledge Google Scholar as the largest bibliometric database; however, we used the second-largest database, Elsevier’s Scopus, because Google Scholar does not provide a bulk export facility for quantitative analysis, and it contains “noise” from non-scholarly references. Secondly, the paper is based on a review so no primary data has been collected or employed. Thirdly, we mapped the literature from the 2000—2022 period but the cadence of innovation in AHRM is accelerating. We acknowledge that 2023 heralded a new type of AI tool in the form of generative AI, such as ChatGPT. Google searches for ChatGPT occurred more than 90 times for the term “generative AI”, and thus, ChatGPT has been voted the “word of the year” for 2023 (The Economist, 2023). Generative AI has much potential to increase the impact of AHRM on two areas identified in this research: recruitment, and learning and development (Budhwar et al. 2023; Rane, 2023;). However, ethical risks arising from bias, fairness, privacy, security, transparency, explicability, and accountability also need to be considered and mitigated (Cameron and Herrmann, 2023; Herrmann, 2023a, b). These undisputable risks that AI applications have in the AI era need to be front and center in future research (Rakowski et al. 2021). We, therefore, propose these aspects of generative AI as a future area of research in AHRM.